xref: /freebsd/sys/powerpc/fpu/fpu_subr.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
1 /*	$NetBSD: fpu_subr.c,v 1.4 2005/12/11 12:18:42 christos Exp $ */
2 
3 /*
4  * SPDX-License-Identifier: BSD-3-Clause
5  *
6  * Copyright (c) 1992, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This software was developed by the Computer Systems Engineering group
10  * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
11  * contributed to Berkeley.
12  *
13  * All advertising materials mentioning features or use of this software
14  * must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Lawrence Berkeley Laboratory.
17  *
18  * Redistribution and use in source and binary forms, with or without
19  * modification, are permitted provided that the following conditions
20  * are met:
21  * 1. Redistributions of source code must retain the above copyright
22  *    notice, this list of conditions and the following disclaimer.
23  * 2. Redistributions in binary form must reproduce the above copyright
24  *    notice, this list of conditions and the following disclaimer in the
25  *    documentation and/or other materials provided with the distribution.
26  * 3. Neither the name of the University nor the names of its contributors
27  *    may be used to endorse or promote products derived from this software
28  *    without specific prior written permission.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40  * SUCH DAMAGE.
41  *
42  *	@(#)fpu_subr.c	8.1 (Berkeley) 6/11/93
43  */
44 
45 /*
46  * FPU subroutines.
47  */
48 
49 #include <sys/cdefs.h>
50 #include <sys/types.h>
51 #include <sys/systm.h>
52 
53 #include <machine/fpu.h>
54 
55 #include <powerpc/fpu/fpu_arith.h>
56 #include <powerpc/fpu/fpu_emu.h>
57 
58 /*
59  * Shift the given number right rsh bits.  Any bits that `fall off' will get
60  * shoved into the sticky field; we return the resulting sticky.  Note that
61  * shifting NaNs is legal (this will never shift all bits out); a NaN's
62  * sticky field is ignored anyway.
63  */
64 int
65 fpu_shr(struct fpn *fp, int rsh)
66 {
67 	u_int m0, m1, m2, m3, s;
68 	int lsh;
69 
70 #ifdef DIAGNOSTIC
71 	if (rsh <= 0 || (fp->fp_class != FPC_NUM && !ISNAN(fp)))
72 		panic("fpu_rightshift 1");
73 #endif
74 
75 	m0 = fp->fp_mant[0];
76 	m1 = fp->fp_mant[1];
77 	m2 = fp->fp_mant[2];
78 	m3 = fp->fp_mant[3];
79 
80 	/* If shifting all the bits out, take a shortcut. */
81 	if (rsh >= FP_NMANT) {
82 #ifdef DIAGNOSTIC
83 		if ((m0 | m1 | m2 | m3) == 0)
84 			panic("fpu_rightshift 2");
85 #endif
86 		fp->fp_mant[0] = 0;
87 		fp->fp_mant[1] = 0;
88 		fp->fp_mant[2] = 0;
89 		fp->fp_mant[3] = 0;
90 #ifdef notdef
91 		if ((m0 | m1 | m2 | m3) == 0)
92 			fp->fp_class = FPC_ZERO;
93 		else
94 #endif
95 			fp->fp_sticky = 1;
96 		return (1);
97 	}
98 
99 	/* Squish out full words. */
100 	s = fp->fp_sticky;
101 	if (rsh >= 32 * 3) {
102 		s |= m3 | m2 | m1;
103 		m3 = m0, m2 = 0, m1 = 0, m0 = 0;
104 	} else if (rsh >= 32 * 2) {
105 		s |= m3 | m2;
106 		m3 = m1, m2 = m0, m1 = 0, m0 = 0;
107 	} else if (rsh >= 32) {
108 		s |= m3;
109 		m3 = m2, m2 = m1, m1 = m0, m0 = 0;
110 	}
111 
112 	/* Handle any remaining partial word. */
113 	if ((rsh &= 31) != 0) {
114 		lsh = 32 - rsh;
115 		s |= m3 << lsh;
116 		m3 = (m3 >> rsh) | (m2 << lsh);
117 		m2 = (m2 >> rsh) | (m1 << lsh);
118 		m1 = (m1 >> rsh) | (m0 << lsh);
119 		m0 >>= rsh;
120 	}
121 	fp->fp_mant[0] = m0;
122 	fp->fp_mant[1] = m1;
123 	fp->fp_mant[2] = m2;
124 	fp->fp_mant[3] = m3;
125 	fp->fp_sticky = s;
126 	return (s);
127 }
128 
129 /*
130  * Force a number to be normal, i.e., make its fraction have all zero
131  * bits before FP_1, then FP_1, then all 1 bits.  This is used for denorms
132  * and (sometimes) for intermediate results.
133  *
134  * Internally, this may use a `supernormal' -- a number whose fp_mant
135  * is greater than or equal to 2.0 -- so as a side effect you can hand it
136  * a supernormal and it will fix it (provided fp->fp_mant[3] == 0).
137  */
138 void
139 fpu_norm(struct fpn *fp)
140 {
141 	u_int m0, m1, m2, m3, top, sup, nrm;
142 	int lsh, rsh, exp;
143 
144 	exp = fp->fp_exp;
145 	m0 = fp->fp_mant[0];
146 	m1 = fp->fp_mant[1];
147 	m2 = fp->fp_mant[2];
148 	m3 = fp->fp_mant[3];
149 
150 	/* Handle severe subnormals with 32-bit moves. */
151 	if (m0 == 0) {
152 		if (m1)
153 			m0 = m1, m1 = m2, m2 = m3, m3 = 0, exp -= 32;
154 		else if (m2)
155 			m0 = m2, m1 = m3, m2 = 0, m3 = 0, exp -= 2 * 32;
156 		else if (m3)
157 			m0 = m3, m1 = 0, m2 = 0, m3 = 0, exp -= 3 * 32;
158 		else {
159 			fp->fp_class = FPC_ZERO;
160 			return;
161 		}
162 	}
163 
164 	/* Now fix any supernormal or remaining subnormal. */
165 	nrm = FP_1;
166 	sup = nrm << 1;
167 	if (m0 >= sup) {
168 		/*
169 		 * We have a supernormal number.  We need to shift it right.
170 		 * We may assume m3==0.
171 		 */
172 		for (rsh = 1, top = m0 >> 1; top >= sup; rsh++)	/* XXX slow */
173 			top >>= 1;
174 		exp += rsh;
175 		lsh = 32 - rsh;
176 		m3 = m2 << lsh;
177 		m2 = (m2 >> rsh) | (m1 << lsh);
178 		m1 = (m1 >> rsh) | (m0 << lsh);
179 		m0 = top;
180 	} else if (m0 < nrm) {
181 		/*
182 		 * We have a regular denorm (a subnormal number), and need
183 		 * to shift it left.
184 		 */
185 		for (lsh = 1, top = m0 << 1; top < nrm; lsh++)	/* XXX slow */
186 			top <<= 1;
187 		exp -= lsh;
188 		rsh = 32 - lsh;
189 		m0 = top | (m1 >> rsh);
190 		m1 = (m1 << lsh) | (m2 >> rsh);
191 		m2 = (m2 << lsh) | (m3 >> rsh);
192 		m3 <<= lsh;
193 	}
194 
195 	fp->fp_exp = exp;
196 	fp->fp_mant[0] = m0;
197 	fp->fp_mant[1] = m1;
198 	fp->fp_mant[2] = m2;
199 	fp->fp_mant[3] = m3;
200 }
201 
202 /*
203  * Concoct a `fresh' Quiet NaN per Appendix N.
204  * As a side effect, we set NV (invalid) for the current exceptions.
205  */
206 struct fpn *
207 fpu_newnan(struct fpemu *fe)
208 {
209 	struct fpn *fp;
210 
211 	fe->fe_cx |= FPSCR_VXSNAN;
212 	fp = &fe->fe_f3;
213 	fp->fp_class = FPC_QNAN;
214 	fp->fp_sign = 0;
215 	fp->fp_mant[0] = FP_1 - 1;
216 	fp->fp_mant[1] = fp->fp_mant[2] = fp->fp_mant[3] = ~0;
217 	DUMPFPN(FPE_REG, fp);
218 	return (fp);
219 }
220